factual

Deep Learning to Compress Detailed Sub-Grid Physics

One promising use of deep learning in cosmology is to run a very detailed simulation of a specific process (like a black hole eating gas) under many conditions, train a model on those results, and use it as a learned set of approximations feeding into the sub-grid—essentially automating and improving the approximations humans currently craft by hand.

factualpending

Speaker

Andrew Pontzen

Evidence Quote

“It would learn some kind of set of approximations to what a black hole is supposed to do, that's what we're trying to do as humans, and then with coding goes in into our sub-grid, it's a way that you can perhaps digest more information”

Source

240 | Andrew Pontzen on Simulations and the Universe— Sean Carroll's Mindscape
Created: 6/13/2026, 12:21:07 AM

Your notes

Loading notes...